How Preinfuse works

The short version

Preinfuse does not test espresso equipment. We collect published specifications from manufacturers, normalise them into one shared vocabulary, and then apply published rules to turn those specifications into fit ratings and setup recommendations. Every fact on the site carries the document it came from and the date we checked it. Every judgement can be traced to the specific facts that produced it.

That is a narrower claim than most equipment sites make. It is also one we can actually stand behind.

What we do not do

We do not physically test equipment. Nobody at Preinfuse has measured an extraction temperature, timed a heat-up, weighed grind retention or tasted a shot for the purposes of this site. You will not find the phrases "we tested", "in our testing" or "after pulling shots" anywhere on it, because they would be false.

We do not republish other people's ratings or reviews. No star ratings, no review counts, no aggregated scores from retailers.

We do not display live prices. Every price on this site is a Preinfuse estimate of the typical US street range, shown as a range and dated. If you need today's price, the retailer has it.

We do not invent social proof. Preinfuse launched in August 2026. There are no testimonials, no user counts and no popularity claims, because there is nothing yet to report.

The three kinds of statement

Everything on this site is one of three things, and we mark which:

Fact. A specification stated by an official source. Recorded with the source, a direct quotation where useful, and the date we checked it. On product pages, these carry a small source marker linking to the document.

Derived. A conclusion computed from recorded facts by a stated method — a unit conversion, a footprint calculation, a budget split. Marked as a Preinfuse derivation rather than a manufacturer statement.

Editorial. A judgement about fit, produced by the published rules below. Never presented as a measurement.

Where facts come from

Sources are ranked, and we use the highest available:

1. Official manufacturer product pages
2. Official instruction manuals and specification sheets
3. Official manufacturer support documentation
4. Authorised distributors, where a manufacturer does not publish for the US market

Where the top sources are silent, we record nothing. A blank on this site means "not published", not "we could not be bothered". That is why some products show fewer specifications than others.

When sources disagree, we do not silently pick one. The conflict is recorded against the specification and shown on the page.

Collection is automated and polite. Every document is fetched once, after checking the site's robots.txt, and cached with its retrieval date and a content hash. We do not scrape retailers, and we do not collect from any source that prohibits it.

Unit normalisation

Manufacturers publish in whatever units suit them. We store one canonical unit per attribute and format for US readers: inches for dimensions, pounds for weight, litres for capacity, watts and volts for electrical.

Where a figure has been converted, the original is shown in the source quotation and the conversion is noted on the row. Lelit publishes the MaraX as 22.5 × 52 ×

35.5 cm; we show 8.9 × 20.5 × 14.0 inches and say so.

How fit ratings are produced

Each product carries up to six ratings on a 1–5 scale. These are not measurements. They are produced by the rules below, applied to that product's recorded specifications. No rating is typed in by hand, which means two products with the same specifications always receive the same rating, and any rating can be opened to see exactly which facts moved it.

Machines

Beginner friendliness — starts at 3, then:

• +1 if grinder is built in, so there is no second purchase to get right
• +1 if milk is textured automatically rather than by hand
• +1 if assisted frothing wand is forgiving to use
• +1 if ships with pressurised baskets, which tolerate an imperfect grind
• +1 if PID keeps brew temperature consistent without your involvement
• -1 if texturing milk by hand is a skill you have to acquire
• -1 if no steam function, so milk drinks need separate equipment
• -2 if a manual lever puts every variable in your hands at once
• -1 if prosumer heating means a warm-up routine and more to understand
• -1 if non-pressurised baskets only, so a good grinder is mandatory from day one

Control depth — starts at 1, then:

• +1 if PID holds brew temperature to a set point
• +1 if brew temperature is adjustable
• +1 if pressure can be shaped across the shot
• +1 if a pressure gauge makes extraction visible
• +1 if milk is textured manually, which is where the control is
• +1 if pre-infusion is under your control
• +1 if a direct lever hands you the whole extraction
• -1 if a super-automatic makes the decisions for you

Milk capability — starts at 1, then:

• +2 if a dedicated steam boiler gives dry steam that does not run out
• +1 if brews and steams at the same time, so drinks do not queue
• +1 if a proper steam wand can produce microfoam, not just froth
• +1 if automatic texturing reaches a consistent result without practice
• -1 if an assisted wand aerates milk rather than texturing it

Workflow speed — starts at 3, then:

• +1 if warms up in under a minute
• +1 if effectively ready on demand
• +1 if grinding happens on the machine, saving a step
• +1 if milk texturing runs unattended
• -2 if needs a long warm-up before the first shot
• -1 if a lever machine is a deliberate, slower ritual
• -1 if brewing and steaming happen in sequence, with a wait between

Upgrade path — starts at 2, then:

• +1 if 58mm-class portafilter opens the full accessory market
• +1 if no integrated grinder, so the grinder can improve independently
• +1 if takes non-pressurised baskets, so better coffee and grinding show up in the cup
• +1 if wear parts and service are realistically available
• +1 if temperature control gives you something to refine
• -1 if a sub-54mm portafilter has a thin accessory market to grow into
• -1 if a super-automatic is replaced rather than improved

Grinders

Beginner friendliness — starts at 3, then:

• +1 if hopper feeding means no weighing before every shot
• +1 if numbered settings are easy to return to
• +1 if doses by weight, removing the most common source of variance
• -1 if grinding by hand takes real effort every time
• -1 if stepless adjustment has no reference numbers to come back to
• -1 if single-dosing adds a weighing step to every shot

Control depth — starts at 1, then:

• +2 if stepless adjustment lets you land between any two settings
• +1 if micro-steps sit inside each macro setting
• +1 if enough settings to find an espresso grind
• +1 if a wide range of settings to work across
• +2 if adjustment steps fine enough for espresso
• +1 if exceptionally fine adjustment steps
• +1 if burrs large enough to grind evenly under load
• +1 if large burrs of a class usually found in commercial grinders
• -1 if not suited to the espresso range

Workflow speed — starts at 3, then:

• +1 if hopper feeding is quicker in the morning
• +1 if doses by weight without a separate scale
• -2 if hand grinding takes about a minute per dose
• -1 if single-dosing means weighing beans for every shot

Upgrade path — starts at 3, then:

• +1 if burr size that will still suit a much better machine
• +1 if burrs and wear parts are catalogued and replaceable
• +1 if stepless adjustment keeps up as your technique improves
• -1 if the espresso range is not where this grinder is strongest

Counter friendliness is derived directly from published width × depth, banded separately for machines and grinders.

How the setup engine works

The setup builder takes four answers — total budget, experience, drinks, priority — and produces up to four complete setups.

Step 1: allocate the budget. A base share for the grinder is set by the total budget, then adjusted for your answers. The share falls as the budget rises, because grinder quality does not scale with price indefinitely while steam capability and temperature stability do.

Adjustments, in percentage points:

The share is clamped between 18% and 48%. A reserve for essential accessories — 8% of the budget, floor $45, ceiling $190 — comes off the top before anything is allocated.

Step 2: rule things out. A machine with no steam function is excluded if you make milk drinks. A manual lever is excluded for beginners. Anything whose lower price bound exceeds the budget is excluded. Exclusions are hard: an excluded product is never scored.

Step 3: score what remains. Each fit rating is multiplied by a weight taken from your priority, then adjusted for your experience and the milk content of your drinks.

Experience and drink adjustments are applied on top of these.

Step 4: pair a grinder. For machines without a grinder, the engine scores every grinder that fits the remaining budget and prefers pairings documented on the product page. A machine with an integrated grinder is scored with a stand-in grinder — maximum convenience, no extra counter space, minimum upgrade path, and a control rating set by its actual number of grind settings — so it competes fairly rather than losing for having fewer parts.

Step 5: penalise ignoring the allocation. A setup whose grinder share drifts more than 8 percentage points from the target loses ground in proportion. This is what stops the engine recommending a $1,900 machine with a $200 grinder.

Step 6: fill the three slots.

• Best overall — the highest-scoring setup that uses at least 55% of the budget.
• Best value — the least expensive setup that retains at least 85% of the best

score. Not the cheapest thing that fits.

• Easiest workflow — the fewest steps in the morning, among setups scoring at

least 80% of the best. If the winner is not genuinely fast, it is labelled "also worth considering" instead.

• Worth stretching for — shown only when a setup up to 20% above your budget

beats the best in-budget option by a clear margin.

Every result is computed at build time and stored. The page you use performs a lookup; it does not re-implement any of this, so what you see cannot disagree with what is written here.

What "best for" means

It means: given the criteria stated on the page, this product fits that use better than the alternatives we list. It does not mean best in the world, best in the category, or best by measurement. Where we say a machine is better for milk drinks, the reason is a specification you can inspect on the same page.

Freshness

Three dates are tracked and shown: when a page was published, when it was last materially reviewed, and when each product's specifications were last verified against source. We do not update a date unless something was actually checked.

Corrections

If something here is wrong, tell us. We would rather fix it than defend it, and corrections to product data are made at the source record so every page that uses it changes at once.

Affiliate relationship

Preinfuse intends to earn commission from retailer links. No affiliate programme is active yet, and no link on this site currently carries a tracking parameter. When that changes, the affiliate disclosure will say so and every commercial link will be disclosed at the point of use.

Retailer links never influence which products appear, how they are rated, or what the setup engine recommends. The engine has no access to commercial data — it scores on specifications and budget only.

Crawler policy {#crawlers}

Preinfuse allows every search and answer-engine crawler full access, because being found is the point. That includes Googlebot, Bingbot, Applebot, DuckDuckBot, OAI-SearchBot, Claude-SearchBot, Claude-User and PerplexityBot.

We also allow model-training crawlers — GPTBot, ClaudeBot, CCBot and Google's extended crawler. This is a deliberate choice rather than an oversight: blocking them would not affect whether Preinfuse is cited in AI answers, which is governed by the search crawlers above, and a new publication benefits more from being known than from withholding text. The decision is recorded in robots.txt as a single block that can be reversed without touching anything else.

We publish an llms.txt file. The current public evidence that answer engines actually read it is weak — the major providers have not committed to it and measured crawler traffic to it is minimal — so we treat it as a low-cost, low-risk convention rather than a growth lever, and we say so rather than implying it does more.